Searching for prior art in the age of AI: The EPO’s approach

A. Aledo Lopez, EPO Chief Operating Officer

Angel Aledo Lopez joined the European Patent Office in January 2019 and was appointed Chief Technology Officer (CTO) in March 2019. In January 2025, he took on the position of Chief Operating Officer, which he combines with his responsibilities as CTO. During his career, he has held different positions in both technical and managerial roles. He has a deep understanding of the IP domain, over two decades of extensive experience in the field, and a strong interest in emerging technologies such as AI and cloud computing.

Artificial intelligence (AI) is reshaping the way prior art is searched and assessed. Commercial platforms increasingly rely on semantic search, vector embeddings, and large language models (LLMs) to improve recall and simplify navigation across vast technical databases. These platforms promise efficiency and broader coverage. Yet for patent professionals, questions remain about reliability, reproducibility, and alignment with the rigorous standards of patent examination.

At the European Patent Office (EPO), AI is being introduced carefully, in a way that combines innovation with legal and procedural rigour, while leaving examiners firmly in control. The flagship example is PreSearch, one of several internal AI-assisted systems that the EPO has developed to support examiners. Around it, additional AI applications are being developed to reduce repetitive tasks and increase the accuracy, consistency, reliability and overall quality of classification, translation, drafting support, and procedural documentation – always under the guiding principle that humans remain responsible for final decisions.

PreSearch: A flagship, AI-powered EPO tool

PreSearch helps examiners identify relevant prior art at the outset of their work. It combines classical retrieval methods with AI, ranking results in a unified list. The system covers both patent literature and non-patent literature (NPL), ensuring broad coverage from the start.

PreSearch integrates a range of algorithms, referred to as “providers,” each offering a different perspective:

Citation providers include references cited by applicants or other patent offices.
Text-based providers locate documents containing distinctive terms from the claims, description or abstract.
AI-based providers add advanced functions such as:
- Vectorisation of patent documents, allowing semantic comparison that goes beyond simple keyword matching.
- Following chains of citations and applying machine learning to highlight relevant prior art.
- Comparing the meaning of claims with earlier patents or scientific articles.
- Generating search markers from claims with the aid of large language models, in alignment with examiner practice.

Planned improvements include a figure-based provider to identify disclosures primarily contained in drawings, as well as enhanced handling of chemical structures and formulae, which is critical for specific technical areas.

Further customised ways of leveraging AI at the EPO

Unlike generic AI solutions, the EPO deploys its own purpose-built models, which are tailored to patent search and examination.

A key example is the custom encoder for pre-classification and classification, trained on millions of patent applications. Because the tool recognises the highly specific terminology and structure of patent documents, it achieves greater accuracy than off-the-shelf algorithms. This allows applications to be routed more efficiently to the correct technical fields and hence supports more precise searches.

Another essential component for examiners is the custom vectorisation of the training database. Before vectorisation, all documents are machine-translated into English, creating a harmonised language base. This ensures that the embeddings capture technical meaning across languages, rather than distorting results due to linguistic variation. The resulting vectors are calibrated for technical and legal relevance, giving examiners cleaner, more focused outputs with fewer irrelevant hits.

Together, these customisations ensure that AI is leveraged at the EPO in a transparent, examiner-oriented manner, designed to strengthen precision, consistency, and trust in the search process.

Benefits of AI in prior art search

The integration of AI into the EPO’s search framework brings several clear advantages. It helps ensure greater consistency in search quality across different technical fields, while also enabling broader coverage by processing and analysing larger and more diverse document collections than was previously feasible. By drawing on a variety of approaches – from semantic methods to classification and other AI techniques – the system supports the identification of higher-quality citations, with documents selected for their actual technical relevance rather than just the presence of certain keywords. At the same time, automated translation allows examiners to conduct truly multilingual searches without losing precision. Together, these benefits strengthen both the efficiency and the reliability of the search process.

AI governance: a human-centric approach

Introducing AI into the prior art search process requires clear safeguards. At the EPO, this is achieved through a well-defined governance framework. All AI outputs are systematically checked against the EPO’s authoritative prior art document collections, ensuring that only documents of verifiable quality can influence the search.

As always, examiners remain firmly in control: AI suggestions may guide the process, but it is always the examiner who decides which documents are relevant and how they are to be cited. Additionally, all search reports and written opinions are checked by the Active Search Division before leaving the EPO.

To keep the tools aligned with real practice, the underlying models are updated regularly, drawing on examiner feedback as well as advances in technology. Most importantly, the EPO follows a principle of human-centric AI. Consistent with the Office’s AI policy, examiners remain responsible for all final decisions, preserving both the legal certainty of the procedure and the professional judgment that only a human expert can provide.

Future developments

The EPO is gradually extending the role of AI beyond prior art retrieval. One area under development is the pre-drafting of office actions, with a pilot planned for the beginning of 2026. Here, AI will assist examiners by suggesting standard formulations and basic legal reasoning, enabling them to concentrate on the substantive analysis of the case.

In addition, since May 2025, the EPO has been piloting AI-assisted minute writing for oral proceedings, an initiative aimed at streamlining documentation and improving procedural efficiency. More broadly, the Office is working towards the gradual introduction of AI assistance across all examiner and formality officer tools, embedding AI support throughout the workflow, while ensuring that humans remain responsible for the outcome.

Together, these initiatives are not designed to replace the expertise of examiners and formalities officers, but to reduce repetitive tasks and allow more time for careful legal and technical evaluation – ensuring that quality remains at the centre of the examination process. This approach is wholly aligned with the EPO’s strategy to simplify and fully digitalise the patent granting process.

Implications for patent attorneys

For European patent attorneys, the gradual integration of AI into the EPO’s processes brings several practical consequences. Search reports and written opinions are expected to become even more consistent, as AI support helps standardise both the way prior art is identified and the format of the written opinions. This improves predictability, as well as giving attorneys and their clients greater clarity on the scope and methodology of searches.

Further, the EPO remains fully committed to maintain dialogue with users. There are numerous channels and opportunities for users to be kept informed and to provide feedback, both with regard to EPO tools and potential legal changes. These include regular user consultation meetings organised by the EPO, which allow attorneys and professional associations to voice their views on new tools, share practical experiences, and help shape how AI is introduced into examination practice.

While these developments promise greater efficiency and reliability, the judgement of examiners and formalities officers remain decisive. For attorneys, this means that while they may benefit from more uniform and predictable outputs, they must still use their own expertise to analyse the reasoning, challenge assumptions and advise clients. In this way, while AI assistance strengthens the framework, ongoing dialogue can continue as previously between attorneys, applicants and the Office.

Conclusion

AI is becoming an indispensable ally in prior art search. At the EPO, its role is carefully defined: AI expands coverage, boosts efficiency, and enhances consistency, while examiners remain the ultimate decision-makers.

Through PreSearch and related initiatives, the EPO demonstrates how AI can support the expertise of examiners and patent attorneys. The result is a search framework that is technologically advanced, legally sound, and examiner-driven, ensuring that EPO prior art searches and examination reports continue to meet the highest standards of quality and reliability.